The UK is positioning itself as an AI leader, and CoreWeave is contributing with two freshly launched GPU-driven datacenters, teeming with Nvidia equipment, now open for business. Situated in Crawley and the London Docklands, these datacenters were developed with Digital Realty and Global Switch, forming a part of CoreWeave’s £1 billion investment initiative, equivalent to $1.2 billion USD.

Despite the hype, these facilities do not feature Nvidia’s latest Blackwell chips, introduced to the market toward the end of 2024. Instead, CoreWeave has opted for the older Hopper architecture GPUs. These datacenters utilize Nvidia’s H200 GPUs coupled with 400 Gb/s Quantum-2 InfiniBand networking, tools from the Hopper generation lauded for their advanced capabilities.

The deployed H200s offer substantial performance with up to 141 GB of swift HBM3e memory and approximately 4 petaFLOPS of sparse FP8 performance per GPU. CoreWeave’s infrastructure design ensures each server is fitted with eight GPUs, a configuration that’s expected to better facilitate the computational demands of modern AI models.

Meta’s Llama 3.1 405B, a front-runner among AI models, can now run fully on a single H200 server – a feat previously requiring distributed nodes or reduced precision on older systems. While CoreWeave has not disclosed the specific number of GPUs per site, its past setups have included clusters exceeding 10,000 accelerators.

These ventures underpin CoreWeave’s ambition to penetrate the UK and broader European markets, reflective of its enhanced £1.75 billion commitment announced during the 2024 International Investment Summit.

As CoreWeave’s global operational footprint expands, including 28 current datacenter locations and 10 additional sites planned by 2025’s close, the firm rides the burgeoning wave of AI-driven market investments.

While CoreWeave paves its way into the UK, the British government’s recent AI Opportunities Action Plan is pressing onward, predicted to yield a £47 billion economic impact by fostering productivity gains through 50 proposed AI strategies.

Yet, this development prompts concerns for the UK’s National Grid, raising questions about the energy infrastructure required to support these expansive AI projects.